MGS does not fail over metrics if the first Nexus returned by DNS is unreachable

During a recent customer incident, one of the three sleds hosting a Nexus zone became unresponsive for an extended period. During this time, no SP metrics from either MGS were captured. Looking at the MGS logs, we observe:

2026-08-24 05:25:13.693Z WARN 51b308da-7500-4aed-beff-ce5035f0ba0e/1905 (producer-server) on oxz_switch: failed to register with Nexus, will retry
    count = 209
    delay = 38215.528565151s
    error = Communication Error: error sending request for url (http://[fd00:1122:3344:10f::24]:12221/metrics/producers): error sending request for url (http://[fd00:1122:3344:10f::24]:12221/metrics/producers): operation timed out
    file = oximeter/producer/src/lib.rs:376

The other two Nexus instances were up and healthy during this time, so MGS would have been able to register with one of them as a producer, had it tried.

Looking at how the Oximeter producer library resolves Nexus, it:

• Calls resolve_nexus_and_register
• This in turn calls out to lookup_socket_v6
• lookup_socket_v6 uses the first entry returned

Using something like lookup_all_socket_v6 and iterating until a healthy Nexus is found would be preferable.

Start in internal-dns/resolver/src/resolver.rs with lookup_socket_v6 and lookup_all_socket_v6, then trace their use from oximeter/producer/src/lib.rs in resolve_nexus_and_register. Verify the producer registration path when the first resolved Nexus is unreachable. Done means MGS can try another returned Nexus and register with a healthy instance.